Different Kinds of Centralization (Hoisted from Comments)

As an addendum to my post about transit cities and centralization, let me explain that the term centralized city really means two different things. One is diffuse centralization throughout the core, typical of pedestrian cities and bus cities and of Paris ex-La Défense; the other is spiky centralization around geographically small transit hubs, for examples Midtown Manhattan, the Chicago Loop, and Central Tokyo. A transit city will tend toward the latter kind of centralization, which is based on walking distance from the subway.

By bus city, I mean a specific kind of urbanism that never existed in the West, but crops up repeatedly elsewhere. It occurs when a city grows too large for walking and cycling while it’s still too poor to build rapid transit, whose construction costs are very high as a share of GDP in developing-world cities. Old buses are not expensive to buy, and their main cost component is labor, which isn’t expensive in a poor city; Beijing for example has only recently gotten rid of conductors on buses.

For a good source on different typologies, I as usual recommend Paul Barter’s thesis – it’s not the main subject of the thesis, but the thesis explains it as background. Bus cities, much like pedestrian cities (which are cities where most people walk to work), tend to be dense all over and monocentric in the sense that there aren’t large suburban centers around them, but they do not have a dominant CBD since buses don’t have the capacity.

Paris is unique among first-world megacities in having preserved this arrangement with its height limits. But it’s still moving in the spiky direction somewhat: the RER has wide stop spacing, which encourages spiky development; and the proposed orbital may be marketed as a circumferential line, but it’s for the most part just a north-south line through La Défense that’s being run together with other lines to potential secondary centers. The difference is that La Défense is more sterile and less pedestrian-friendly than Midtown Manhattan and the Chicago Loop. I may write about this in another post, but greenfield CBDs seem to be always worse for pedestrians than legacy ones, and if the legacy CBD hasn’t evolved to the spiky transit city form, then urbanists may conflate the spiky transit city form with the pedestrian-unfriendliness of the greenfield CBD.

Transit city centralization works differently – it’s based on walking distance from the main rapid transit nodes. Recall that transfers at the downtown end are the most inconvenient for suburban commuters, so that one subway stop away from the center is too far. This makes the transit city CBD inherently geographically small, so that the job density is much higher than that of any other urban form; the job density can also be higher because of the larger amount of space afforded by skyscrapers.

In contrast, the transit city is unlikely to be monocentric. A dominant CBD accessed by rapid transit is a geography that tends to create extremely long commutes – much longer than car-accessible edgeless cities, though not longer than trying to access the same CBD by car – and this leads governments to promote the growth of secondary centers, which are also spiky. Because those secondary centers look like CBDs and not like endless sprawl as do the secondary centers in the US, they make the city look polycentric, even if measured in terms of the CBD’s share of metro area employment they’re very CBD-dominated. When I say a transit city is inherently a centralized city, I do not mean that secondary centers are impossible or undesirable, just that the CBD needs to have a relatively large share of jobs, and that the secondary centers should be actual centers – if they can’t be like Shinjuku, they should be like Jamaica or Newark or how Tysons Corner wants to look in 20 years and not like how Tysons Corner looks now.

For example of how this kind of centralization emerges from the other kind, we can look at the evolution of cities that built large rapid transit networks. Tokyo around Nihonbashi would be the best example, but New York around City Hall is as good. While Lower Manhattan is clearly a smaller CBD than Midtown, it still looks like a spiky CBD, which it did not a hundred years ago. If you plot the locations of the skyscrapers in Lower Manhattan, with few exceptions they’re all south of Chambers, usually far south; peak employment is around Fulton and Wall Streets. The old elevated terminal for Brooklyn trains at Park Row would be inappropriately located. North of Chambers there are city neighborhoods with names like Chinatown or TriBeCa, which are mixed-use enough to have many jobs but have nowhere near the job density of Wall Street.

A related kind of centralization occurs in a multipolar city region, composed of many small cities. None of the cities of the Ruhr is large enough to spawn spiky subcenters on its own, but because the region has grown so interdependent it’s as big as a megacity, the legacy centers in the various cities have turned into a spiky centralization, only without one CBD dominating the rest.

I think it’s the last kind of spiky centralization that transit advocates think of when they propose to turn LA into a multipolar region. Or perhaps it’s in a limbo between a true multipolar region and a unipolar one with well-defined, transit-oriented secondary CBDs. On the one hand, the transit lines proposed in and beyond Measure R are not very downtown-centric. Each direction out of downtown generally gets one line, the exception being the west because of the low-hanging Expo Line fruit and the higher-demand Wilshire corridor. The focus is on connectivity between different poles, since unlike a true transit city Los Angeles has no capacity crunch on its transit system. The subway proposal for going beyond Measure R is to continue south of Wilshire on Vermont, missing downtown entirely, rather than, say, continuing east of Union Station along Whittier.

But on the other hand, the secondary cores are defined in relation to downtown – west (Santa Monica, UCLA), north (Burbank), south (Long Beach), and so on. It’s not like the organic buildup of agglomeration that merged the various cities of the Ruhr into one megaregion, or the merger of the metro areas of New York and Newark, or on a larger scale San Francisco and San Jose. Instead, these secondary cores emerged as secondary to Downtown LA, and only became big because Downtown LA’s transportation capacity is limited by the lack of rapid transit. Put another way, a transit revival in Los Angeles that includes rapid transit construction would make Los Angeles more downtown-oriented rather than less.

A Transit City is a Centralized City

In New York, a large fraction of employment clusters in a rectangle bounded roughly by 59th Street, 2nd Avenue, 42nd Street, and 9th Avenue. Although it’s a commonplace that New York employment is centralized around Manhattan, in reality most of Manhattan is residential, and employment is concentrated in a few square kilometers in the heart of Midtown. This is where the subway lines converge from all directions – elsewhere there simply isn’t enough capacity. Of course it wasn’t always like this: Manhattan’s population in the 1890s was the same as it is today, and it was clustered toward the southern third of the island, but employment was relatively evenly distributed in the downtown area. What has happened since then is that New York became a transit city.

There’s a strong correlation between the form of a city and the mix of transportation options people use. This extends well beyond density, but the principle is the same. Transit is at its best at high intensity, because this is what supports high-frequency service. Cars are the opposite: even on a normal urban street, a car alone will beat any rapid transit line, but every additional car will slow down the road dramatically, so that at even the moderate intensity of an edge city gridlock ensues.

Although usually this principle is stated in terms of density, it’s equally true for work centralization. The pedestrian city and the bus city will be dense all over, and feature high job density scattered across neighborhoods: walking is too slow for the transit city pattern to emerge, and buses have too little capacity. But dedicated rapid transit wants to serve an area right next to the stations, and once a network is built, a CBD grows around the central area. This CBD is typically small, just a few square kilometers. Even vaguely CBD-ish locations, such as Penn Station, are too far, as one commonly quoted figure about work locations demonstrates. The CBD isn’t even large enough to encompass all of the 34h-59th Street strip that the tourist guidebooks define as Midtown. The subway lines only form a tight mesh in a subset of that general area.

The job density of such a CBD is measured in hundreds of thousands per square kilometers, requiring many high-rise towers, several of which are supertall. In contrast, most of New York’s residences are mid-rise, and Tokyo’s are low- and mid-rise; their residential densities in the low tens of thousands per square kilometer are high enough that they are considered the epitome of density, but their CBDs are an order of magnitude denser.

Of the major transit cities of the world, Paris is the only one that’s resisted this trend with its height limit, but instead a transit-like CBD started out in La Défense, and the same pattern that comes from the subway in New York or Tokyo or the L in Chicago emerges with the RER. Of course, Paris maintains very high residential density, but its job distribution is more in line with that of a bus city – employment is dense all over, and the Downtown Paris employment density peak is less pronounced than in comparable transit city downtowns.

This does not mean a transit city needs to have empty trains going in the reverse-peak direction, as Cap’n Transit, Jarrett Walker, and others charge. A transit city will have job destinations outside the CBD, growing around rapid transit junctions: for example, Tokyo has Shinjuku, Shibuya, and Ikebukuro, all of which are so replete with high-rises it’s hard easy to forget they’re secondary job centers. While there is still a pronounced peak direction, people rely on transit so much that they take it for regular errands, supporting very high off-peak frequency by the standards of trains with drivers.

New York has something similar in Downtown Brooklyn, Jamaica, and Long Island City, but the modal split of those job destinations is much less favorable to transit – 50% in Downtown Brooklyn and Long Island City and 30% in Jamaica, according to a study of New York’s secondary job centers that I can no longer find. This is a general feature of many old American cities: the core looks like a transit city, but beyond it is a car-centric city, filled with edge cities and edgeless cities. Because the layout beyond the core is car-centric, the off-peak and reverse-peak traffic that supports high all-day bidirectional frequency on the Tokyo rail network, or for that matter on most New York City Subway lines, does not exist. The preference of American commuter rail agencies for peak-only service comes partly from an operating model that makes it impossible to run frequent off- and reverse-peak service, but also from a job distribution that makes the market for such runs small even under the best industry practice.

A corollary of this fact is that the multipolarity of other cities, for example Los Angeles, is not an asset. It would be an asset if those job centers were intense and could be easily served by transit; in reality, they have moderate intensity, nothing like that of the secondary centers of Tokyo or even New York, and serving many of them requires digging new subway lines. Burbank, on the legacy Metrolink network, could make a reasonable site for a transit-oriented secondary center, if commuter rail operations were modernized and local transit lines were extended to it; the Westside and Santa Monica do not, and the hope is that the investment in the Subway to the Sea could enable them to grow to reasonable size.

The key here is that the reason Shinjuku, Ikebukuro, and Shibuya are as transit-oriented as Central Tokyo is that they historically arose as connection points between the Yamanote Line and the private railroads. In particular, they already had rapid transit fanning out from multiple directions when they became major job centers. But Tokyo’s transit development history is peculiar; most other cities did not have large electrified rapid transit systems terminating at the edge of the urban core prior to building local subway lines.

A second corollary then is a strategy that sought to make New York a more transit-oriented city would treat centralization differently. It should turn the secondary centers into transit nodes in their own right, with tails extending as far out as reasonably possible. Jamaica already has some of the infrastructure, but it’s used poorly because of antiquated LIRR practices; the same can’t be said of Flushing, so a priority should be to build reasonable-quality transit from multiple directions, connecting Flushing with College Point and Jamaica and modernizing the LIRR so that it could connect it with Bayside.

A point that many people writing about this neglect (with pleasant exceptions like Cap’n Transit, the Streetsblog crowd, and Paul Barter) is that this requires both the carrot of more transit and the stick of less parking. In any case it’s hard to create high job densities when much of the land is used for parking. But on top of that parking mandates make it difficult for transit to be competitive when it’s expected to include railyards and depots in its budget and roads are not.

But what a transit city doesn’t need is job dispersal. The importance of creating secondary centers is strictly as alternatives to auto-oriented edge cities and edgeless cities, since whatever happens, not all jobs will be in the CBD. A large city with rapid transit connecting to all major neighborhoods will automatically have high transportation capacity. Rapid transit is good at transporting tens of thousands of people in one direction in the peak hour; let it do what it’s good at.

New York-New Rochelle Metro-North-HSR Compatibility

Let me preface this post by saying that there should not be any high-speed trains between New York and New Rochelle, except perhaps right at the northern end of the segment. However, to provide reasonable speeds from New York to Boston, it’s desirable to upgrade the maximum speed between New York and New Rochelle to 200 km/h or not much less. The subject of this post is how this can be accommodated while also permitting some regional rail service, as proposed by the MTA. There are two reasons to bundle the two. First, some of the work required could be shared: for example, new stations could be done at the same time as rail and tie replacement. And second, the presence of both upgraded intercity rail and regional rail on the line requires some four-tracking and schedule optimization.

The physical infrastructure required for boosting speeds within New York City is fairly minimal by itself. The right-of-way in the Bronx has some curves but they are not very sharp and can be somewhat straightened without knocking down buildings, and even the curves in Queens and on the Hell Gate Bridge, while unfixable without major viaduct modification, are not terrible if superelevation is high and tilting is enabled.

A big question mark is what the maximum speed permitted by the physical layout of the East River Tunnels is. Current speed is 97 km/h (60 mph), but top speed today in other sections of the network are below those achieved decades ago (for example on Portal Bridge), and trains with specially designed noses, as the Shinkansen rolling stock is, could potentially go even faster. Regardless, it is not important for HSR-regional rail integration, since the East River Tunnels have no stops and will be running far under capacity once East Side Access opens. Thus, all travel times in this post are between New Rochelle and Sunnyside Junction, which is notionally considered to be located at 39th Street. This is a 25-kilometer segment.

Another question mark is what the speed limit on the S-curve south of New Rochelle is. Currently the limit is 48 km/h (30 mph). Raising it requires grade-separating the junction between the NEC and the current New Haven Line. It can be raised further via curve straightening, but the question is how much eminent domain can be done. The maximum radius that can be achieved with minimal or no eminent domain is 700-800 meters. Some further eminent domain may be required to have this curve start far enough from the southbound platform that full 200 mm superelevation is achievable without subjecting local train riders to too much cant excess. For comparison, slicing through New Rochelle and the Pelham Country Club allows essentially eliminating the curve and allowing maximum speed through the area, which taking surrounding curves into consideration is about 240 km/h.

Assuming 150 km/h (about 700 meters radius, 200 mm cant, and 175 mm cant deficiency), the technical travel time for a nonstop intercity train between when it passes New Rochelle and when it passes Sunnyside is about 9 minutes; this includes slowdowns in Queens and the Bronx and on Hell Gate. A nonstop M8 with a top speed of 145 km/h would do the same trip in about 11:15. (Amtrak’s current travel time from New York to New Rochelle is about 25 minutes, of which by my observation riding Regional trains 6 are south of Sunnyside.)

Even the above travel time figures require some four-tracking, independently of capacity, in order to limit cant excess. Unlike the Providence Line, the Hell Gate line has some curves right at potential station locations – for example, the Hunts Point stop is located very close to the curve around the Bruckner Expressway, and the Morris Park stop is located in the middle of a curve. The Bruckner curve radius is about 500 meters, and 200 mm superelevation would impose 80 mm cant excess on even a fast-accelerating commuter train (1 m/s^2 to 72 km/h), and an uncomfortable 140 mm on a slower-accelerating one (0.5 m/s^2 to 51 km/h). The Morris Park curve is even worse, since it would impose a full 200 mm cant excess on a stopped train. So we should assume four-tracking at least at the Morris Park station, which is located in the middle of a curve, and Hunts Points, and potentially also at Parkchester.

Now, a local train would be stopping at New Rochelle and four stops in the Bronx, and should be stopping at Sunnyside. Although a FLIRT loses only about 75 seconds from a stop in 160 km/h territory, assuming 30-second dwell times, the M8 is a heavier, slower-accelerating train, and for our purposes we should assume a 90-second stop penalty. This means that, counting New Rochelle and Sunnyside together as a single dwell-free stop (they involve one acceleration and one deceleration in the Sunnyside-New Rochelle segment), local technical travel time is 18:15, about the same as what Amtrak achieves today without stops but with less superelevaiton and inferior rolling stock.

Now, 18:15-9 = 9:15, 9:15 times the schedule pad factor is 9:54, and modern signaling allows 2-minute headways up to 200 km/h; thus we can accommodate 4 tph intercity and 4 tph local Metro-North without overtakes except at New Rochelle and Sunnyside.

There is only one problem with the no-overtake scenario: the MTA plans on a peak traffic much higher than 4 tph, in line with the New Haven Line’s high demand. It’s planning on a peak of 6-8 tph according to what I’ve read in comments on Second Avenue Sagas. This naturally breaks into 4 tph that make local stops and 4 that do not (though my suspicion of MTA practice is that it wants fewer than 4 local tph); if there are fewer than 8 trains, one slot could be eliminated.

Let’s look then at a 4/4/4 scenario. Assume that trains depart Sunnyside in order of speed – HSR first (passing rather than stopping at Sunnyside), then express Metro-North, then local Metro-North. A local train will be overtaken first by the following HSR, and then by the following express. If we could move the overtake point to New Rochelle, the local would not need to wait for trains to pass it. In reality, 4/4/4 means the local departs Sunnyside 4-5 minutes after the HSR train passes it, and has 9 minutes of time penalty before being overtaken again. If the stop penalty could be reduced to 75 seconds, then the overtake could be moved to New Rochelle, demonstrating the use of top-quality rolling stock. But the M8s are good enough for many purposes, and therefore we will not assume a noncompliant replacement, unlike in the case of the MBTA, whose rolling stock is slow and very heavy.

With 9 minutes of time to make up, it’s tempting to have an overtake at a four-tracked Co-op City station. But then the local would have to be overtaken by two trains in a row, and moreover the two trains would become quite separated by then due to differing top speeds, and this would force a penalty on the order of 6 minutes.

I claim that the best would be to four-track a segment between two or even three stations; the right-of-way is wide enough anyway. In addition, the Morris Park curve could be straightened if the Eastchester Avenue overpass were modified, and doing this in conjunction with four-tracking would be cheaper than doing each alone. Under this option, the local would leave Sunnyside much later than 2 minutes after the express, just enough to be overtaken by HSR at Morris Park. It would then keep going to Co-op City until overtaken by an express. This would essentially save about 2.5 minutes out of the 6 in penalty, since the train would be in motion for that time.

New Rochelle-Penn Station Regional Rail

Last week, the MTA again floated proposals for connecting Metro-North to Penn Station once East Side Access comes online and frees track space currently used by the LIRR. The New Haven Line is to be connected to Penn Station via Northeast Corridor trackage that only Amtrak uses today, with four new stations in the Bronx. The new station locations include one near Co-op City, a dense middle-class housing project that is underserved by transit, and three more neighborhoods that are inconveniently between the 5 and 6 or on the wrong side of a freeway. In sum, it is a positive development.

However, since the New Haven Line already has a Manhattan terminal at Grand Central, this project involves splitting the line in two in its inner section. Thus, frequency will be cut in half, unless there is extra service added north of the merge point at New Rochelle. At the peak, this is not a very big problem, since the New Haven Line runs 20 trains into Grand Central between 8 and 9 am every weekday; although this is misleading since most stations are only served by a small subset of these trains, it is not difficult to have trains make a few more stops to restore the existing frequencies.

The problem is off-peak service. The current pattern is one train per hour serving stations north of Stamford and running nonstop between Stamford and Harlem-125th, and two Grand Central-Stamford locals per hour in the weekday off-peak and one on weekends. While poor by any international standards, the service afforded to the lower New Haven Line is tied for best in the US with just a handful of lines with half-hourly off-peak service. Splitting frequency in half would be a disaster for such service, to say nothing of not being useful to regional riders in the Bronx. Moreover, adding service just so that it can be split south of New Rochelle is counterproductive: the greatest need for frequency is close to the center rather than in the suburbs, because the shorter the trip time, the more pronounced the effect of a long wait time is.

I claim that the best way to compromise on frequency under the current service paradigm is to run short-turning trains terminating at New Rochelle, with timed connections. Since some passengers prefer a one-seat ride, half the local trains should serve Penn Station and half should serve Grand Central. In other words, frequency should be split among the two Manhattan destinations, but each branch should have a short-turning train connecting with the other branch’s trains. Express trains should make a station stop at New Rochelle with a reasonable connection from the local trains, but should otherwise only serve one destination. Then, Grand Central is the better destination for express trains, since it minimizes interference with intercity trains.

The alternative is to turn New Rochelle-Penn Station into a modernized regional rail line, run somewhat independently of the rest of Metro-North, with through-trains from the rest of the New Haven Line only at rush hour. Maybe select few off-peak trains, no more than 1 per hour, could extend to Stamford. This requires a change in paradigm; it cannot be done with the current staffing levels or turnaround times, but since it’s a service expansion, it’s plausible if unlikely that the union will accept reduced staffing, in line with best practices.

I envision the following scenario for modernized regional rail:

– Trains go from New Rochelle to Penn Station and beyond, to New Jersey. Through-service to the Hudson Line via the Empire Connection avoids agency turf battles but is less useful for passengers. They can hook into existing services and go all the way to Trenton and Long Branch, or provide new service and only go as far as Newark.

– Minimum off-peak frequency is one train every 15-20 minutes, or perhaps 30 late at night. 10 is aspirational, if the service proves popular.

– Fares are integrated with local transit. This means intra-city trips cost the same as subway or not much more, and in either case, transfers to the subway or the buses are free. If people can ride trains and a ferry from Tottenville to Wakefield on one fare, people should get to ride direct from Co-op City to Penn Station on one fare.

– Trains make stops that interface with other transit options. A Sunnyside stop meeting with the LIRR is a must. In addition, if the grades permit, there should be a stop in Astoria meeting the subway, and perhaps one in Port Morris, so that the trains can offer fast frequent service between Queens and the Bronx. Perhaps there should also be a restored station meeting buses from City Island.

The Sunnyside stop has value no matter what: for one, it allows trains to Penn Station to also work as Grand Central trains, making the transferring process easier to implement. The other extra stops are not really useful unless commuter rail is made an attractive option for local trips – in short, an S-Bahn or RER rather than a traditional American commuter service.

I hope to discuss compatibility with modernized intercity trains tomorrow. Although half-hourly service is so infrequent there is no real interference with intercity trains, more frequent service could pose problems. This is not an issue if Amtrak is not modernized: the speed limit south of New Rochelle is at most 160 km/h and even that is only between the Hutchinson River and Pelham Manor, with 100-110 km/h on the rest of the line. Thus the only speed difference between regional and intercity trains comes from making station stops, and a glance at existing schedules shows that when the top speed is 130-140 km/h trains lose about 1.5 minutes per stop. Of course high-powered noncompliant trains lose much less time, but for the purposes of running punctually on a shared line, the M8s are good enough. Losing 6 minutes from the four planned station stops is not a problem even with the proposed peak frequency, once one remembers that most peak trains are not going to stop in the Bronx at all.

More Track Maps

A kind reader sent me the two maps on Rich E Green’s now-offline website that I did not have, namely maps of all of Connecticut and Rhode Island. These join earlier maps I’d posted of the Northeast Corridor in Maryland, and the commuter railroads in the Mid-Atlantic, separated into Long Island, Metro-North and Empire South, and New Jersey Transit and SEPTA.

Update: based on request by the author, I took down all the maps, and scrubbed the links.

Compromising with Agency Turf

Daniel Krause added his two cents to the politicals vs. technicals issue; his contention is that technical advocates are perfectionists and refuse to compromise. Writing about the Transbay Terminal design, which is slightly less wretched than originally planned but still severely constrains Caltrain capacity into downtown San Francisco, he says,

-It seems [technicals] have a difficulty in accepting design compromises. The Transbay Terminal situation is a good example. Even though the turning radius has been significantly widened, it is still tighter than they like, so the drumbeat continues about how bad and non-function the Transbay Terminal will be, even though (in my opinion) a reasonable level of service will now be accommodated within the terminal for both Caltrain and HSR.

-Things they find as deal-breakers in terms of design, are not deal-breakers to the general public. The public just wants a comfortable and reasonably fast way to get around. For example, they will not be bothered if trains go a little slower into the Transbay terminal because the turning radius is not ideal. They will still be perfectly happy to arrive into the heart of downtown SF despite the ongoing debate in the blogosphere.

The telltale sign that something is wrong is that there’s nobody to compromise with on this matter. Compromises based on community needs happen all the time. In fact one of my future projects for this blog is to propose an elevated alignment in Providence from the East Side Rail Tunnel to downtown, and this involves a series of compromises between cost, noise, takings, and speed. The problem is that the compromises leading to the Transbay design, or, worse, the design of San Jose Diridon, are not based on any local needs; they’re based on the needs of agencies that won’t cooperate. The same is true of the various caverns built or proposed in New York, and Harold Interlocking.

Although I’ve voiced the view that experts should be thought as one more constituency with its own special knowledge (namely, best industry practices), they should not be viewed as a constituency with its own interests. They should serve the public, not the reverse. And the public should pressure them to come up with designs that maximize passenger convenience.

In the case of Transbay specifically, the agency turf is not just leading to high cost. The worst aspect of it is that most peak Caltrain trains will not be able to serve Transbay, but instead have to terminate at the present 4th and King terminal, separated from the CBD by a kilometer and a pedestrian-hostile freeway connector. Passengers will be forced to transfer to the Central Subway, a very slow and low-capacity line that isn’t expected to get much more ridership than the buses it’s replacing.

Although it’s tempting to view passengers as automatons who only care about glossy trains, the reality is that the little details matter.  If the area around a station isn’t walkable, people will not walk to it. If the timetable is too inscrutable, or schedule reliability is poor, or the trains squeal, passengers will be more likely to look for alternatives. Of course laypeople may not be able to tell exactly what is wrong – they may complain that transferring is horrible but not know that transfers could be made cross-platform and timed, or they may complain about paying a series of fares but not know that fare unions exist – but they can tell something is wrong. Ordinary people are much less stupid than the elites think they are. The best argument against democracy may be five minutes’ conversation with the average voter, but the best argument for democracy is five minutes’ conversation with the average member of the elite.

I for one was not born hostile to American transit agencies. I became this way after, first, riding Amtrak; and second, stumbling across a cost per km figure for Tokyo subways that was much lower than the equivalent New York figure, eventually leading to my interest in comparative costs. And judging by the deteriorating position of HSR in the polls in California, the design incompetence is having a similar effect on many others.

Trust (Hoisted from Comments)

Robert Cruickshank’s much-anticipated reply to my posts about political versus technical transit supporters and their activism says that high-speed rail is a political issue, and therefore what’s important is to just get it done.

To me, the problem comes from my unfortunate choice of the terms political and technical. The main difference is not about technical concerns; it’s about whether one trusts American transit agencies. Thus I don’t really see the point when Robert complains about neo-liberalism and the evils of financial cost-benefit calculations. The terminology I picked may have reinforced the image of technicals as heartless engineers and technocrats, but in reality the opposite is true. Technicals have a much bigger standard deviation in their political attitudes than politicals; they range from Rothbardian libertarians to free speech advocates and people who make fun of the phrase “undisclosed location” in the context of US-sponsored torture. The common thread is mistrust of agency officials; the technical arguments are there because when we disagree with officials rather than just report what they say, we need to actually rebut their claims.

In contrast with Robert’s picture of the technical as a technocrat, my technical activism comes from the opposite end: it’s a rejection of a self-justifying bureaucracy that equates “build nothing” with “continue to build highways” and that thinks progress equals megaprojects. It’s a matter of supporting consensus politics and informed citizenry rather than subservience to agency officials. US government officials spend 2-10 times more on infrastructure projects as they have to. They have agency turf battles that make transit less user-friendly, and to cover up those turf battles they propose to spend billions of dollars on gratuitous viaducts, caverns, tunnels, and what have you. They write passenger rail-hostile regulations. And when called on it, they defraud the public and even tell outright lies. Trust in government agencies is so low that when the California HSR Authority admitted to the cost overruns, the LA Times treated it as a moment of honesty.

It’s precisely this trust that people care about, and it’s eroding when HSR becomes the equivalent of $600 toilet seats. Of course there is money for transit, but it’s either wasted or not given to transit because people can’t trust that it can be used wisely. I view it as part of my goal to showcase how good transit can be done, so that it doesn’t look so expensive for the benefit provided.

A fundamental tenet of risk perception theory is that people are most concerned about risks they find morally reprehensible – and this collusion between government and government contractors offends me. Just because it’s greenwashed doesn’t mean it’s any better than subsidizing oil drilling, paying military contractors $1,000 per day, or bailing out financial companies that then use the money to pay the executives who caused the financial crisis multi-million dollar bonuses. No wonder that when Republicans talk about the ingenuity of individual business leaders, they talk about Mark Zuckerberg, the Google guys, and Steve Jobs; they have to go that far out of the industries that give money to the GOP, such as oil, to find people who’ve actually innovated rather than just sucked public money. In fact one of the impetuses for the spread of neo-liberal boosterism in popular culture is the perception that entrepreneurs who are untainted by the public sector are good, while government is inherently incompetent and corrupt. When the government doesn’t do a good job, people stop believing it’s even possible for good government to exist.

Yonah Freemark writes that it doesn’t matter if costs are high because HSR costs are a small part of the transportation budget, which is itself a tiny part of GDP. But transportation is also not the biggest priority in spending. Most of the GDP, even most government spending, is and should be things that aren’t transportation; and most transportation funding isn’t and shouldn’t be intercity.

For an order of magnitude of what other issues are involved, Robert is proposing $1 trillion in student loan forgiveness as economic stimulus. My point is not to impugn him; I agree with him there. It’s that the big-ticket items are not transportation, but instead transportation is one of many small-ticket items of spending. But pool many small expenses – a hundred billion here, a hundred billion there – and you’re starting to talk about real money.

And this is true politically, not just economically. The Democratic Party has been advocating for universal health care since the Truman administration. After early successes with Medicare and Medicaid, its efforts stalled; its empathy-based appeals went nowhere. In Politics Lost, Joe Klein writes about how Bob Shrum would insert the phrase “health care is a right, not a privilege” into the speeches of every Presidential candidate he worked for – and how every candidate he worked for lost. Meanwhile, US health care costs were ballooning faster than those of other first-world countries. By 2005-6 it was impossible to miss, and liberal pundits seized and owned the issue, portraying American health care as not only inequitable but also inefficient. Five years later, they got their universal health care bill, flawed as it is. Nowadays the people who are pooh-poohing the idea of health care cost control are Greg Mankiw and the Tea Party.

Spending is a zero sum game, but economically and politically. The Great Recession won’t last forever. Any infrastructure building plan is going to outlast the recession, triggering real tax hikes, spending cuts, or interest rate hikes in the future. It’s fine if the infrastructure is cost-effective; it’s not fine if it isn’t. (In comments on CAHSR Blog, I was told that the example of Japan shows that the recession can last forever; if it does, the US will have bigger problems than transportation.)

And this is equally true politically. The amount of government spending is controlled tightly by the political acceptability of deficits. Some deficits are more politically acceptable than others – for example, military waste is acceptable to many right-wingers – but in this political climate, HSR is at least as controversial on the right as extending jobless benefits, and far less useful as stimulus per dollar spent. The unemployed tend not to fork over much of their benefits to international consultants. If a few billion dollars are enough to showcase workable HSR then by all means the administration should spend them, but if they’d eat $20 billion out of a $50 billion jobs bill that Obama’s going to run for reelection on, there’s no point.

I think that both on transportation and on health care, there’s a political not-invented-here reasons among the partisans. Liberals owned health care cost control, so Greg Mankiw started arguing that it wouldn’t help society much and that high costs are a good thing and Sarah Palin referred to cost control as death panels. The issue with transportation is a little different; while many technicals are leftists, it’s anti-urban conservatives and Koch-libertarians who cancel transit projects, use phrases like “the money tree,” and demagogue about how no rail project is ever affordable. My instinct is to point out that those conservatives have no trouble overspending on road projects and rationalizing highway cost overruns; but if you think in terms of spending, and treat transportation as one program of many stimulus projects, there’s a real not-invented-here issue here.

Ironically, despite Robert’s claim that costs don’t matter and benefits do, much of what I rail against is exactly benefits. I personally am reminded by how awful the turf battles are every time I have to buy an MBTA ticket at the cafe since Amtrak bullied the MBTA out of the Providence station booths, and every time I take the subway to Penn Station and need to change concourses to get my Amtrak ticket. The key for me is to make transit cheap enough that it can be deployed on a large scale, and to make it convenient and pedestrian-friendly, which park-and-ride-oriented commuter rail is not.

Do Not Compare NEC with HSR Ridership

One common claim doubting high ridership estimates for American high-speed rail lines is that the Northeast Corridor gets little ridership. For example, commenter Gelboak says,

How plausible is a 51-77 million px / year ridership?

I believe the NE corridor currently has a ridership of approximately 10 million px / year, and I think the CAHSR and the NE Corridor have not all that different population sizes in the respective catchment areas. And many of generally the metro areas in the NE Corridor have larger shares of their population in dense urban cores than is the case in the Bay Area, Sacramento, LA/Orange County & San Diego.

This line has also been repeated elsewhere, among serious urban economists like Ed Glaeser as well as among commenters and journalists. It is a bad comparison, for one reason: the Northeast Corridor does not have high-speed rail. The Regional averages 110 km/h between New York and Washington and 85 km/h between New York and Boston; the premium-priced Acela averages 130 and 100 km/h respectively. This is the speed of mildly upgraded legacy lines – faster than the lines full-fat HSR replaced in Japan and France, much slower than actual HSR or heavily upgraded lines in Germany and Britain. It’s not much faster than a bus, which is why Megabus and Bolt have been so successful in the Northeast, while no comparable service exists on Tokyo-Osaka or Paris-Lyon.

In contrast, express Shinkansen trains average 200 km/h or more on the three main Shinkansen routes. So do TGVs running between cities served by LGVs, without needing to use legacy lines except in immediate station areas. For example, the fastest bi-hourly TGVs from Paris to Marseille average 230 km/h. Upgraded legacy lines are also fast by NEC standards: Britain’s East Coast Main Line and West Coast Main Line trains from London to the major cities in Scotland and the north of England usually average 140 km/h, and in one case (London-York) close to 160 km/h. For a selection of fastest-train speeds in Europe, see DoDo’s analysis at EuroTrib.

If we look at ridership on full-fat HSR, we get much higher numbers than NEC ridership. The Tokaido Shinkansen  had 141 million passengers in fiscal 2010. The Sanyo Shinkansen gets 60 million passengers per year. The LGV Sud-Est and Atlantique each got 20 million annual passengers within 10 years of opening and the Madrid-Seville AVE line 6 million (see p. 109 of the new CAHSR business plan) , and although I have no link at hand, I’ve read the LGV Sud-Est is up to 30 million by now. Taiwan HSR and the KTX both significantly underperformed expectations, but each gets about 35 million annual passengers now.

If we measure the populations of the cities connected, the Northeast Corridor is easily the busiest potential HSR line in the West. New York is larger than any other Western city, and nowhere else in the first world is such a city connected to three second cities as large as Philadelphia, Boston, and Washington. Using the KTX as an analogy, New York and Seoul have about the same metro area populations, each of Washington and Boston is about the same size as Busan or a little larger, and the smaller intermediate cities, such as Baltimore and Providence, are about the same size as Daegu and Daejeon. The NY-DC and NY-Boston distances are slightly shorter than Seoul-Busan, reducing HSR’s advantage over the highways, but conversely the KTX is quite slow, averaging about 170 km/h as of 2009. The ridership on the NEC may be 10 million with today’s service levels, but it should be far well above 50 million with higher speeds and good punctuality.

Together with the fact that international links underperform, the result is that the two lines most frequently used by critics as examples of low ridership – Eurostar and the Acela – are not comparable to the HSR systems proposed. Thus sanguine expectations of ridership are realistic, issues with connecting transit and low road tolls aside. The Northeast Corridor competes with congested, tolled roads and feeds into cities with regional rail systems that suck except for when they need to serve the Amtrak station; it is not the example one should use to rebut American HSR proposals. Let an American HSR line open first before criticizing.

The upshot is that although the costs of HSR in the US have blown out of proportion, the benefits are still high. Maybe not high enough to cover the costs, but higher than the critics think. Agency turf battles and general contractor incompetence give high-cost, high-maintenance projects, but they don’t make passengers not want to ride the lines.

Quick Note: What Does Profitability Mean?

The 2012 business plan for California HSR, in addition to admitting to wanton cost overruns, also promises that the system will be profitable. Or does it? I did not want to comment on the plan’s notion of profits, but I see via California HSR Blog that several outfits have seized upon that part and treat the release as much better news than it actually is.

The plan says, e.g. on page 15 of the PDF, that the system will generate operating profits even on the lowest-ridership scenarios. This has led Treehugger and others to crow that this will not be a disaster. But a careful consideration suggests the opposite. The medium scenario posits $1 billion in revenue in 2025 versus $539 million in annual operating costs. But those operating costs exclude depreciation; by then the project is expected to expend about $50 billion, which at even a mild depreciation schedule is more than enough to put the system in the red.

The problem is that people in the US are used to judging transit and rail profits using transit agency metrics, by which other people pay for capital and therefore the main operating ratio excludes depreciation. This is not normal accounting; EBITDA is a much less important metric than EBIT (including depreciation but not interest) or net income (including everything) for a normal, profitable business. The profitability of HSR outside the US is measured in terms of net profits; in Taiwan, the system has had positive EBITDA since a few months after opening, but went bankrupt due to elevated interest charges.

The argument that the business plan proves something special because of the positive EBITDA may satisfy people who get their criticism of HSR from hacks who conflate capital and operating costs, but it should not satisfy people who occasionally bother to read railroad budgets. The higher the quality of a line, the lower the operating costs are excluding depreciation and the higher the depreciation and interest charges are. For example, see this breakdown of Madrid-Barcelona HSR costs and profits; infrastructure charges are dominated by depreciation and interest rather than maintenance, though rolling stock charges are more maintenance than depreciation.

Even state-of-the-art HSR infrastructure maintenance is cheap. The 2012 business plan a little more than $100,000 per route-km (cf. €30,000 per single track-km according to a 2008 study, which works out to about the same modulo inflation and a high Euro:dollar exchange rate). It’s a second-order term. The same is true of avoidable operating costs, such as rolling stock maintenance and labor. Of course ten second-order terms make a first-order term, and indeed the total operating costs of HSR are not negligible. However, they’re still lower than depreciation charges.

The importance of including depreciation is that HSR capital doesn’t last forever. Rolling stock has to be replaced. Viaducts and tunnels need to be refurbished. It’s hard to come up with exact figures since HSR lines have not yet depreciated in full in the 47-year history of the technology, but railroads all over the world have accountants who include depreciation terms in the budget. Of course, the problem is that if the capital cost is too high, then the depreciation and interest will weigh the project down. This hasn’t really been observed abroad, except in cases in which the interest rates were very high as in Taiwan, but judging by the business plan’s numbers, it could happen in California.

Finally, although the biggest bombshell in the plan is the cost overrun, the plan also has a ridership shortfall. It’s not a big shortfall, but on page 115 the plan mentions that the revised full-buildout ridership estimate for 2035 is 51-77 million, depending on fares, down from 69-98 million according to the 2008 environmental impact statement. This partly explains why the operating revenues are so low relative to full operating costs including depreciation.

The CAHSR Bombshell

The 2012 CAHSR business plan has some bombshell construction cost numbers: the headline number is $98 billion, leading to predictable complaints that the cost has run over by a factor of 3 over the original $33 billion budget of 2008. This is somewhat misleading since it includes inflation, but there’s still a factor-of-2 real cost overrun to investigate: in 2010 dollars the cost is $65 billion, as predicted by CARRD though with a somewhat different distribution of cost overrun among the various segments.

Some of it is scope creep that could be removed later via value engineering, and some is additional delays. The new plan assumes construction will take until 2033, vs. 2020 originally. The one point of light is that the initial construction segment (ICS) from Fresno to Bakersfield is still within budget, giving time to send the people involved in scope creep to early retirement and do the designs better. The biggest cost overruns are on the Peninsula and LA Basin segments, which are now up to $25 billion, about triple the original cost estimate. This already suggests that lack of money is what is causing costs to grow: just as it’s expensive to be poor, so is it expensive for an agency to have no money and drag construction over decades, in many segments.

But it’s not just the delays. The Peninsula blended plan includes many extra features, such as $1.5 billion for 80 km of electrification (in Auckland the same amount of electrification cost $80 million), $1 billion for 10 km of very tall and unnecessary viaducts through downtown San Jose, and $500 million $1.9 billion to tunnel under Millbrae (see update below) in order to preserve BART’s three tracks.

There’s scope creep and there’s scope creep. Sometimes, a project’s costs go up because new features are added that are useful (for example, converting a single-track diesel project into a dual-track electrified light rail, as was done on the LA Blue Line), or that are necessary but were glossed over initially in order to keep cost estimates down. A little bit of the latter kind of scope creep is present in the Central Valley, in the form of more viaducts than originally planned; CARRD’s cost overrun estimate was based entirely on taking CAHSR’s unit costs and applying them to the added features as of 1-2 years ago. But the kind of scope creep we see on the Peninsula is entirely different: they are adding features that are of marginal operational use, and instead exist mainly to reinforce agency turf lines (namely, separation of agencies at San Jose).

My suspicion is that the same is true of the other segments. The fact that a cost overrun was averted on the initial construction segment in the Central Valley, after extensive value-engineering (for example, fewer viaducts), shows that the one segment CAHSR needs to build within budget in order to survive is indeed being built within budget. The other segments, for which the HSR Authority hopes to obtain private and local funding, offer easy opportunities for contractor profiteering: once the initial segment is built, there may well be momentum to complete the system, and the consultants could strong-arm local governments and the federal government to cough up more money. Indeed, no extra features useful to passengers have been added – everything is just about agency turf and more viaducts.

The only places where there could plausibly be an honest overrun, which cannot be eliminated simply by putting adults in charge and going back to older plans, are the mountain crossings. And indeed, the Grapevine alternative, now posited to be $1-4 billion cheaper than the Tehachapis, could resolve the major issue heading south toward the LA Basin. In the north, they keep studying the Altamont overlay with options including one proposed by SETEC that lets trains run at full speed right up until the built-up area of southern Alameda County; together with the Dumbarton water tunnel, it could help the project stay within budget by switching to a superior alternative, and avoid the San Jose viaduct mess entirely.

Although the political supporters of CAHSR tend to discount the Grapevine and be skeptical of switching to Altamont, they are still interested in the option of value-engineering. But it’s stupid to first propose an outrageous plan and then value-engineer it back to the original cost estimate. It offers no political advantages over doing it right the first time, and just breeds justifiable mistrust of the authority. For all I know, there could be a large real overrun that is not the result of agency turf wars.

To make sure people don’t react to the apparent factor-of-three overrun the way they should – i.e. propose to pull the plug unless costs are scaled down to reasonable levels – the 2012 plan includes higher numbers for the cost of doing nothing, i.e. of expanding freeways and airports to provide the same capacity. It was originally $100 billion, and is now $170 billion. This is less self-serving than it seems: the plan assumes a slower buildout and higher inflation, which accounts for most of the difference. But it’s still a backhanded way of trying to force the state to kick more money toward the contractors. If they can slow down airport and freeway construction (thereby increasing the final cost), perhaps they can halt it entirely – fair’s fair.

I’m still optimistic that they could put adults in charge and reduce costs to the original estimate, as they already have in the Central Valley. That is, if the federal government dangles a few billion dollars for the LA-Bakersfield segment and demands even a modicum of accountability, then they will gladly use the money to build a useful initial operable segment and only try to extort the public later. But optimistic and certain are not the same, and it’s an outrage that such a project could cost $65 billion. The tunnel-heavy Shin-Aomori extension of the Tohoku Shinkansen cost $4.6 billion for 82 km, a little more than half the proposed per-km cost of the new business plan – and Japan is a high-construction cost country.

Unless they cut the costs, I don’t see how I can continue to support the project. The initial construction segment, useless as it is on its own, is fine; the question is whether it stakes the territory for a very expensive future extension, or for one with reasonable cost. Since I doubt they’ll be able to get any additional money until they connect to the LA Basin except from the federal government and even then it will be a small number of billions, I think it’s the latter option. But the rest should be scrapped and restarted unless the construction costs drop dramatically. I would peg the maximum that the project can cost before it should be canceled, on the outside, at $60 billion or so in today’s money. This assumes timely construction – waiting decades with rapidly depreciating track hosting limited service makes the situation worse. The only consolation I have is that no matter what, the other projects they could spend the money on if CAHSR is canceled are even worse. And this says more about those other projects than about CAHSR.

Update: here is the cost escalation breakdown. It’s overwhelmingly the addition of new features, i.e. tunnels and viaducts, most of which are unnecessary (though one major issue, additional tunnels from Palmdale to LA, is required due to further study showing the need for more environmental protection). For example, Millbrae gets a gratuitous tunnel, previously estimated at $500 million, now estimated at $1.9 billion (p. 20). Unsurprisingly, SF-SJ has the biggest overrun, a factor of 2.5. Hat-tip goes to Clem for noting the extra cost of Millbrae, which I missed looking at just the business plan.

Update update: the California HSR Authority links rotted away, but were replaced with new ones. The page references remain valid; the reference to the cost in the first link at the beginning of this post is PDF-p. 15, and the reference to the breakdown of cost overrun by segment is in the update link, PDF-pp. 7-10. The cost estimate for the project was since revised down to $53 billion, in 2011 dollars, in the final 2012 business plan (see PDF-p. 23); this is entirely from leaving out the LA-Anaheim and SF-SJ segments for later, which avoids the Millbrae tunnel and other Peninsula luxuries, but does not address the extra costs of going through Palmdale or the cost overrun just south of San Jose.